以人工免疫法建構管制圖非隨機性模型分類架構之研究

Implementation of Artificial Immune System Applied to Nonrandom Patterns of Control Charts

葉宗遠(Tsung-Yuan Yeh)

指導教授 ： 陳雲岫

元智大學/工程學院/工業工程與管理學系/碩士(2004年)

摘要

在統計製程管制中，管制圖是用來管制製程的品質以及研判製程的狀態。當製程在管制中發生可歸屬原因之影響時，可經由辨識管制圖中的非隨機性模型，繼而採取製程改善的措施。本研究主要應用人工免疫法具備多樣性辨識與演算進化的能力，以MATLAB程式撰寫人工免疫系統的分類演算法，並建構一個管制圖非隨機性模型分類之架構，透過演算法所演化出之抗體來辨識正常模型、畸形模型、平均值偏移模型、趨勢模型、層別型模型、混合型模型、週期性模型、系統性模型以及雙群體模型等九種管制圖之模型。
本研究驗證結果顯示，人工免疫法所建構之分類架構，能有效地辨識管制圖之異常模型，其中分類效果最顯著的有畸形模型、層別型模型、混合型模型、週期性模型、系統性模型以及雙群體模型，而正常模型分類效果有限，且容易與平均值偏移或趨勢這兩種模型混淆。本研究另建構一小型輔助分類架構，採用兩階段分類方式，可改善原始架構在上述三種模型特性上之分類限制。
本論文研究成果能提供管制圖重要且有價值的資訊，幫助管理者瞭解製程變異的可能原因並提早採取適當的改善方案，以避免擬定錯誤的改善策略或製程不適當地被調整，而導致更大的製程變異與成本損失。

關鍵字

統計製程管制 ； 管制圖 ； 非隨機性模型 ； 人工免疫法 ； 分類

並列摘要

The control charts of statistical process control（SPC）are used to monitor the status of the process and to detect any situations which are out of control. The nonrandom patterns of control charts indicate the process affected by assignable causes, and we should take a corrective action to improve the process immediately. In this project, we construct an immune system framework for nonrandom patterns of control charts to differentiate and classify the nine different patterns via the immune classification algorithm, which are normal, freaks, sudden shift in level, trends, stratification, mixtures, cyclic, systematic and grouping（or called “bunching”）patterns. An immune algorithm is formed and executed by the MATLAB.
The results of the study show that the immune classification algorithm could differentiate the nonrandom patterns of control charts. The types of freaks, trends, stratification, mixtures, cyclic, systematic and grouping are recognized significantly while normal, sudden shift in level and trends are failure to be identification as well as the other six types.
The immunity-based classification framework will be expected to effectively judge the kind of unusual model of control charts in time, and to analyze the random and nonrandom causes of the exceptional manufacturing procedure. That will provide worthy information of control charts to the manufacturing procedure, and the decision maker will promote the quality of products correctly in the early stage to avoid making incorrect improvement strategies or adjusting the manufacturing procedure inappropriately which induces more exceptions of the manufacturing procedure and the loss of the cost.

並列關鍵字

statistical process control (SPC) ； control chart ； nonrandom patterns ； artificial immune system (AIS) ； classification

參考文獻

7.楊素芬（2002），品質管理，華泰，台北。

2.林裕章 (1992），類神經網路應用於統計製程管制非隨機性模型之研判，元智大學工業工程與管理研究所碩士論文。

3.馬誠韋 (2001)，解答多目標規劃的新方法－免疫系統法，元智大學工業工程與管理研究所碩士論文。

8.Chen, C. S. (1997), “A Neural Network Approach for the Analysis of Control Chart Patterns,” International Journal of Production Research, 35, 667-697.

9.Dasgupta, D. and Attoh-Okine Nii (1997), “Immunity-Based Systems: A Survey,” IEEE proceeding, 369-374.